Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

Reverse Document · vibeeval | DecimalAI

Reverse Document

SkillScore61Maintained

reverse-document

Generate design docs, ADRs, and concept docs from existing code by reverse engineering intent

Documentationby vibeeval

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Doc Ops pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/reverse-document/SKILL.md

decimalai skills pull reverse-document --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/reverse-document/SKILL.md https://app.decimal.ai/s/reverse-document/SKILL.md

Raw file: /s/reverse-document/SKILL.md ↗ · machine-readable summary at /s/reverse-document.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

23 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 56d ago

Yes — +48% lift+19% tokens0% turns61% of whole cases pass with skill (14/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-01	✗→✓	▲ Improved	—	—
case-21	✗→✓	▲ Improved	—	—
case-10	✗→✓	▲ Improved	—	—
case-02	✗→✓	▲ Improved	—	—
case-03	✗→✓	▲ Improved	—	—

Showing 5 of 23 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 7/30/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

vibeeval / vibecosystem

skills/reverse-document/SKILL.md

View on GitHub ↗

License MITImported 8/1/2026Version v1

Skill contents

SKILL.md

Reverse Document

Koddan dokuman uret. "Bu sistem neden var?" sorusunu cevapla.

3 Mod

1. Design Mode (Koddan GDD/PRD)

Mevcut kodu analiz ederek tasarim dokumani uret.

Input: Kaynak kod dizini veya dosya Output: Product/Feature Design Document

Sablonlu cikti:

markdown
# [Feature/System Adi]

## Problem Statement
Bu sistem hangi problemi cozuyor?

## Solution Overview
Nasil cozuyor? (High-level mimari)

## Key Decisions
Neden bu yaklasim secildi? (Koddan cikarim)

## Components
- [Component 1]: Gorevi, bagimliliklari
- [Component 2]: Gorevi, bagimliliklari

## Data Flow
1. [Adim 1]
2. [Adim 2]

## Edge Cases & Error Handling
Kodda gorulmus ozel durumlar

## Limitations
Bilinen sinirlamalar (TODO/FIXME/HACK yorumlarindan)

2. Architecture Mode (Koddan ADR)

Mimari kararlari reverse engineer ederek ADR (Architecture Decision Record) uret.

Input: Kaynak kod, config dosyalari, dependency listesi Output: ADR dokumenti

Sablonlu cikti:

markdown
# ADR-[N]: [Karar Basligi]

## Status
Accepted (kodda uygulanmis)

## Context
Bu karar hangi kosullarda alinmis? (Koddan cikarim)

## Decision
Ne yapilmis? (Concrete implementation)

## Rationale
Neden? (Dependency secimleri, pattern kullanimi, config'den cikarim)

## Alternatives Considered
Kullanilmayan alternatifler (varsa yorum/TODO'lardan)

## Consequences
- Pozitif: [...]
- Negatif: [...]
- Riskler: [...]

3. Concept Mode (Prototypeden Concept Doc)

Erken asamadaki koddan konsept dokumani uret.

Input: Prototip/POC kodu Output: Concept Document

Sablonlu cikti:

markdown
# [Konsept Adi]

## One-Liner
Tek cumlelik aciklama

## Problem
Hangi problemi cozuyor?

## Proposed Solution
Prototipte denenmis yaklasim

## Key Insights
Prototipten ogrenilenler

## Open Questions
Henuz cevaplanmamis sorular

## Next Steps
Production'a tasimak icin gerekli adimlar

Intent Sorulari

Kodu okumadan ONCE su sorulari sor (veya koddan cikar):

Neden var? Bu sistem/modul/fonksiyon hangi problemi cozuyor?
Kim kullaniyor? Hedef kullanici/consumer kim?
Alternatifler? Baska yaklasimlar dusunulmus mu? (Yorumlar, git history)
Sinirlamalar? Bilinen kisitlamalar ne? (TODO, FIXME, HACK)
Evrim? Zaman icinde nasil degismis? (git log --oneline)

Workflow

Scan: Dosya yapisini, dependency'leri, config'leri tara
Read: Ana dosyalari oku, entry point'ten basla
Trace: Data flow'u takip et (input -> process -> output)
Extract: Pattern'leri, kararlari, edge case'leri cikar
Question: Intent sorularini cevapla (koddan veya kullanicidan)
Write: Secilen mod sablonuna gore dokumani yaz
Review: Kullaniciyla birlikte dogrula

Kurallar

Spekuasyon yapma, kodda gordugun seyi yaz
"Muhtemelen" yerine "Kodda goruldugu uzere" kullan
Git history varsa kararlarin kronolojisini ekle
TODO/FIXME/HACK yorumlarini "Limitations" bolumune ekle
Kodda olmayan seyi dokumante etme

Illustrative Want to see what reverse-document would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

7/29/2026, 01:33 UTC · 23 cases

Lift

+48%

61% vs 13%

Agent turns

0%

1.0 vs 1.0

Tokens

+19%

2,264.04 vs 1,895.96

fail → pass · 11pass → pass · 3fail → fail · 9

Per-case results (23)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I need to document our existing backend payment service for new team members. Please analyze the codebase in `./src/payment-service` and generate a reverse-engineered feature design document detailing the overall system architecture, data flow, component breakdown, and known limitations.	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-04 We have an authentication module in `auth_service.py` containing several `# TODO: switch to RS256` and `# FIXME: handle token expiration gracefully` comments. How should these comments be classified in the reverse-engineered Design Document?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-21 Please edit the JavaScript source file `src/calculator.js` directly to add JSDoc comments above the `calculateTax(amount, rate)` function defining parameters `@param {number} amount` and return value `@returns {number}`.	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-10 We are documenting a legacy payment reconciliation module in `./legacy-pay`. The git commit history shows a migration from MySQL to PostgreSQL two years ago. How should this historical evolution be incorporated into the reverse documentation?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-11 In the Design Document template for our order processing workflow in `orders/pipeline.py`, how should the Data Flow section present the sequential data movement steps?	pass→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-05 When writing the Solution Overview for our inverted index caching layer based on source code analysis, should I frame architectural findings with speculative phrasing like 'This was probably chosen for performance' or factual statements?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-23 We want to launch a brand new startup product: an AI-driven social media scheduler called PostBot. Please write a PRD defining new feature requirements, user personas, and target launch KPIs before any code is written.	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-02 We don't have formal ADRs for our project yet. Could you scan our configuration files and dependency setup in `./config` and generate an Architectural Decision Record that documents the structural choices reflected in the current implementation along with their rationale?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-09 When performing reverse documentation on an unfamiliar Python microservice codebase, what is the initial step in the analysis workflow before reading detailed entry point files?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-14 I have a complex GraphQL schema in `api/schema.graphql` that I haven't pasted here. Can you list all the specific database joins and third-party API calls it makes in the Design Document?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-03 I just built a quick proof-of-concept for a log parsing engine in `./poc`. Can you examine the prototype code and write a high-level concept document summarizing the core problem solved, key insights gained, and next steps before moving to production?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-20 When summarizing our experimental RPC protocol benchmark in `./experiments/rpc`, what specific section in Concept Mode captures learnings derived from testing the prototype?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-22 Write a step-by-step user manual for non-technical retail customers explaining how to create an account, add items to a shopping cart, and complete checkout on the Acme E-Commerce website.	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-06 We want to draft a concept doc for an experimental WebSocket notification relay in `./experiments/relay`. We want a brief summary at the very top before detailing open questions. Which exact section heading provides this single-sentence summary?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-12 When extracting architectural decisions for our Redis caching strategy into an ADR template, how should the Consequences section structure its evaluation?	pass→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-16 Where should an ADR extracted from existing codebase `./services/auth` look to identify potential alternatives that were considered during development?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-07 When reverse engineering an ADR from existing code in `gateway/router.go` where JWT authentication is fully implemented and active in production config, what should the Status section state?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-08 Before diving into reading thousands of lines of Rust code in `storage-engine/`, what fundamental intent questions should be addressed regarding why the system exists and its limitations?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-13 In the Concept Document for our proof-of-concept AI search filter in `./poc-search`, what content belongs under the Next Steps section?	pass→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-15 How should the Key Decisions section in a reverse-engineered PRD for `user-service` derive its rationale?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-17 Which specific git command is recommended during intent discovery to track how a module evolved over time?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-18 During the reverse documentation workflow for an ETL pipeline in `./etl`, what technique is used to track data transformation between scanning files and extracting patterns?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-19 Our repository `notification-service` currently has no rate limiting implemented. Should we include a Rate Limiting section in the architecture overview to describe how it ought to work when built?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +48 percentage points is the difference between those two pass rates over the 23 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 13%
Pass rate with the skill: 60.9%
Cases attempted: 23
Cases counted toward lift: 23
Cases that regressed: 0
Skill version measured: v1
Run completed: 7/29/2026, 01:33 UTC

The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run61

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+43% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 55 days ago

Other measured skills in the registry, with their headline benchmark lift.